Online Optimization of Difference-of-Convex Compositions with Smooth Mappings

Jingwei Ji; Jong-Shi Pang; Renyuan Xu

REVIEW 2 major objections 4 minor 106 references

Online Optimization of Difference-of-Convex Compositions with Smooth Mappings

T0 review · 2 major / 4 minor · reviewed 2026-08-01 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read The paper shows that a proximal residual computed from convex subproblems certifies first-order stationarity for online nonconvex optimization with difference-of-convex composite losses and constraints.

desk verdict A genuine extension of local-regret to CDCS objectives with CDCS constraints, but the stationarity certificate depends on a Slater condition that is not verified in the examples; worth refereeing after revision. read the letter →

arxiv 2607.19553 v1 pith:WT5J4URP submitted 2026-07-21 math.OC cs.LG

Jingwei Ji , Jong-Shi Pang , Renyuan Xu This is my paper · ORCID

classification math.OCcs.LG MSC 90C2690C2590C30

keywords onlineoptimizationdifference-of-convexnonconvexnonsmoothproximalresidualstationaritymeasuretangentconelocalregretconstraintqualification

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper studies online learning where each round's loss is a difference-of-convex function composed with a smooth map, and where the feasible region is itself defined by a constraint of the same kind. Because generic nonconvex problems are intractable, the paper adopts a stationarity-based benchmark rather than global optimality. Its central claim is that a time-smoothed proximal-linear algorithm, whose inner updates are convex subproblems, has a natural residual that is a proper stationarity measure: whenever the residual's fixed-point condition holds at a point, that point satisfies a first-order stationarity condition for the original nonconvex constrained problem. The key enabling result is a tangent-cone characterization showing that a carefully cushioned linearization of the constraint set preserves the local geometry of the true feasible set. If correct, the paper delivers a local-regret bound, a bound on the number of inner convex solves, and an error bound that turns the residual into a computable certificate of approximate stationarity.

What carries the argument

The cushioned linearization set L̃(x): a convex set formed by linearizing the inner smooth map ζ at x inside the convex function φ, linearizing the entire composite concave part ψ∘ζ, and adding quadratic cushion terms ±½Lβ‖y−x‖² to both sides. Theorem 1 says that under the CDCS Slater condition, T_{L̃(x)}(x)=T_X(x)=C(x), where C(x) is the cone of directions d satisfying φ'(ζ(x);Jζ(x)d) ≤ ∇ψ(ζ(x))ᵀJζ(x)d. This tangent-cone equality is what allows the proximal-linear mapping S_t^μ(x), a convex problem over L̃(x), to certify stationarity of the original nonconvex problem. The proximal residual G_t^μ(x) is the scaled displacement produced by this mapping.

What would settle it

Build a two-constraint CDCS example where both constraints are active at a boundary point so condition (28) has no solution, run the algorithm, and check whether a fixed point x̄=S_t^μ(x̄;v) still satisfies 0∈b∂Φ_t(x̄)+T_X(x̄)^*. If a fixed point occurs without the stationarity inclusion, that example marks exactly where the Slater assumption is load-bearing.

Watch

Extended reading notes

Core claim

On its own terms, the paper's discovery is that the proximal-residual mapping G_t^μ(x)=μ(S_t^μ(x)−x), defined as the difference between the current point and the minimizer of a convex proximal-linear surrogate, is a faithful first-order stationarity measure for the nonconvex constrained problem. Proposition 2 proves that the fixed-point condition x=S_t^μ(x;v) implies 0∈b∂Φ_t(x)+T_X(x)^*, which is Bouligand stationarity of the time-smoothed loss over the original feasible set. The mechanism is Theorem 1: under a CDCS Slater condition, the tangent cone of the convex surrogate set equals the tangent cone of the true feasible set, so moving between the surrogate and the original problem loses no

Load-bearing premise

At every boundary point of the feasible region where the constraint is active, there must exist a direction along which the convex part grows strictly slower than the linearized concave part; without such a Slater-type direction, the cushioned surrogate set may fail to reproduce the true tangent cone, and the stationarity certificate can break.

Editorial extensions

If this is right

The local-regret bound Regret_T ≲ T(δ/w+(L_f+L_g)L_θ/w)² says the iterates are approximately stationary on average, with per-round deviation controlled by the stopping tolerance and the smoothing window.
Each inner iteration solves a convex optimization problem, so the method runs with a convex oracle even though the objective and constraints are nonconvex.
The bound on total inner iterations, O(Mμ²w²/(μ−(L_f+L_g)β_θ)δ⁻²(1+T/w)), keeps the computational overhead finite and polynomial in the precision requirement.
Feeding the same loss at every round recovers an offline method: suitable choices of window and tolerance yield an iterate whose expected squared residual is below any ε, i.e., a first-order stationary point.
Under additional regularity, Proposition 4's error bound makes the residual a quantitative certificate: a small residual implies the next iterate is close to the true stationarity set.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

Not in the paper: the tangent-cone characterization is likely reusable for offline optimization with CDCS constraints, where it would justify convex-surrogate algorithms without a separate stationarity analysis.
Not in the paper: when the feasible region is a max of several CDCS constraints, the single-constraint Slater condition must be checked for that max representation; verifying or relaxing it for multiple simultaneously active constraints would extend the framework to common constraint structures.
Not in the paper: replacing the uniform sliding window with adaptive or exponentially weighted smoothing should preserve the proof's logic, since the regret argument only needs the smoothed objectives to change by a bounded amount between rounds.
Not in the paper: the residual could serve as a practical stopping criterion in stochastic or bandit-like online settings where only partial loss information is available at decision time.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Short version: this is a real extension, not a repackaging. It is the first to put CDCS objectives together with CDCS constraints in the online local-regret framework, and the tangent-cone characterization (Theorem 1) is a new technical tool. The proximal residual is a sensible stationarity measure when the CDCS Slater condition holds: the proof of Theorem 1 is coherent, and Proposition 2 follows from it. The regret and iteration-complexity bounds in Theorem 2 are standard but correct, given the Lipschitz/smoothness assumptions. I would send this to referees.

The soft spots are real. First, the CDCS Slater condition (Definition 2) is load-bearing. A bounded variant of the stress-test example makes this concrete: take X=[0,∞) via c(x)=|ζ(x)|−ζ(x) with ζ=x^3/(1+x^4). At x=0, Jζ(0)=0 so no d0 satisfies the strict inequality; the cushioned set collapses to {0}, T_{~L}(0)≠T_X(0), and for Φ(x)=−x the residual is zero at 0 while 0 is not stationary. The stress-test note as written uses ζ=x^3/(1+x^2) and says the range is bounded, which is wrong because c is unbounded, but the bounded variant makes the same point. The paper does not verify Slater for Example 4; the Huber-ℓ_0 constraint appears to fail it at saturated boundary points. Second, Proposition 4's proof uses β_g and β_ψ smoothness of ∇g_s and ∇ψ that are not in Assumption 1 or the favorable condition. That is fixable by adding assumptions, but as written the error-bound claim is not fully supported. Third, the paper does not discuss what happens when Slater fails: the residual may still be a reasonable heuristic, but it is not a certified stationarity measure, and the local-regret bound loses its interpretation.

I do not think these are fatal. The main framework and the tangent-cone geometry are sound, and the missing pieces are identifiable rather than hidden. But the paper should not be accepted as is: the Slater condition needs to be checked for the examples, and Proposition 4's hypotheses need to be stated. A serious editor should send this to peer review, with referees asked to focus on Slater verification and the error-bound assumptions.

Referee Report

2 major / 4 minor

Summary. The paper proposes an online optimization framework for nonconvex nonsmooth losses of CDCS form f_t(θ_t(x))−g_t(θ_t(x)) over a nonconvex feasible set X={x: φ(ζ(x))−ψ(ζ(x))≤0}, where φ,ψ are convex, ψ,g are differentiable, and θ,ζ are smooth. The core algorithmic idea is a time-smoothed proximal-linear method whose inner subproblems are convex over a "cushioned linearization" set L̃(·). The paper defines a proximal residual G_t^μ(x)=μ(S_t^μ(x)−x) and a local-regret measure Σ∥G_t^μ(x_t)∥². Its main theoretical claims are: (i) Theorem 1, a tangent-cone characterization T_{L̃(x̄)}=T_X(x̄)=C(x̄) under a CDCS Slater CQ; (ii) Proposition 2, stating that a fixed point of S_t^μ implies first-order stationarity 0∈∂̂Φ_t(x̄)+T_X(x̄)*; (iii) Theorem 2, giving a local-regret bound of order T(δ/w+(L_f+L_g)L_θ/w)² and an inner-iteration bound; and (iv) Proposition 4, an error bound relating ∥G_t^μ(x)∥ to dist(0,∂̂Φ_t(x⁺)+T_X(x⁺)*). The proofs of the regret and iteration bounds are largely self-contained and appear coherent.

Significance. If the stated results hold, this would be a meaningful extension of the local-regret framework to a broad structured nonconvex nonsmooth class with nonconvex constraints, while keeping all inner updates convex. The tangent-cone characterization in Theorem 1 is of independent interest for offline CDCS-constrained optimization. The main regret bound has the expected scaling and the iteration-complexity analysis is nontrivial. However, the significance is currently weakened by two substantial gaps: the stationarity certificate depends on a CDCS Slater condition that is not verified for the paper's own examples and is not automatic even for simple constraints; and Proposition 4 relies on smoothness assumptions on ∇g and ∇ψ that are not stated in the paper's assumptions. These gaps affect the central claim that the residual is a proper stationarity measure, and one of the paper's advertised contributions.

major comments (2)

[§3.1, Definition 2 and Theorem 1, eq. (28)] The CDCS Slater CQ (28) is load-bearing for the equality T_{L̃(x̄)}=T_X(x̄) and hence for Proposition 2, but it is not automatic and is not verified for the examples. Counterexample: let X=[0,∞) be represented by φ(t)=|t|, ψ(t)=t, ζ(x)=x³/(1+x²). All favorable-condition assumptions hold. At x̄=0, Jζ(0)=0, so no d0 satisfies φ'(0;0)<0; Slater fails. Then L̃(0)={0}, so T_{L̃}(0)={0}, but T_X(0)=[0,∞). With Φ(x)=−x, S(0)=0 and G(0)=0, yet 0∉∂̂Φ(0)+T_X(0)*=(−∞,−1]. Thus the residual is not a proper stationarity measure without Slater. Moreover, Example 4 (the Huber ℓ0 budget constraint, eq. (17)) fails (28) at every active point: at such a point the zero coordinates are exactly zero and the large coordinates have zero derivative, so φ'(x;d)−∇ψ(x)ᵀd = Σ_{j:x_j=0}|d_j| ≥ 0 for all d. The abstract's unqualified claim that the residual is a proper stationarity measure is therefore too broad; the
[§3.5, Proposition 4, eqs. (95) and (99)] The proof of Proposition 4 invokes Lipschitz constants β_g and β_ψ for ∇g_s and ∇ψ, respectively. Neither Assumption 1 nor the favorable condition (Definition 1) includes smoothness of ∇g or ∇ψ; only L-Lipschitzness and differentiability of g and ψ are assumed. In bounding ∥∇g_s(θ_s(x⁺))−∇g_s(θ_s(x))∥ and the analogous term for ∇ψ, the proof uses exactly these unstated constants. Consequently, the error bound (81) is not established under the assumptions stated in the proposition. The proposition should be restated with explicit smoothness assumptions on ∇g_s and ∇ψ (and then these should be verified for the examples, e.g., the Huber function does have a Lipschitz gradient), or the claim should be modified to avoid these terms.

minor comments (4)

[§3.5, eq. (81)] The displayed bound (81) has malformed typesetting with stray '+ +' and missing parentheses, making it hard to parse; it should be corrected.
[Abstract and §1.1] The phrase 'proper stationarity measure for the original problem' should be qualified by the CDCS Slater condition, since Proposition 2 explicitly depends on it.
[Example 4, eqs. (14)–(17)] The constraint (15) is only sufficient, not equivalent, to the ℓ0 budget (14); the paper says 'it suffices,' but the subsequent discussion of feasible regions should not imply that (15) exactly characterizes the sparsity constraint.
[§1.3] The reference 'Zhu and Li (2026)' cites arXiv:2607.08954 with a July 2026 date; please verify that this reference is complete and accurately described.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the stationarity certificate and regret bound are derived from stated assumptions, not from fitted values or load-bearing self-citations.

full rationale

The paper's central claim is that the proximal residual G_t^mu(x)=mu(S_t^mu(x)-x) is a proper stationarity measure. This is proved in Proposition 2 via Theorem 1, which characterizes T_X(bar x) as C(bar x) under the CDCS Slater condition. The proof of Theorem 1 is self-contained: it starts from the definitions of the cushioned linearization set L~, the candidate cone C(bar x), and the Bouligand tangent cone, and derives the equality by explicit inequalities and a constructed in-set sequence. No fitted parameter is renamed as a prediction, and no external benchmark is defined in terms of the algorithm's output. The local-regret bound in Theorem 2 follows directly from the algorithm's stopping rule (residual below delta/w) plus a Lipschitz drift bound between consecutive smoothed objectives; it is a proof of a property of the proposed update, not a fitted or self-referential prediction. The paper does cite works co-authored by one of the current authors (Cui and Pang 2021; Pang et al. 2017), but these citations are used for terminology, standard DC calculus, and context, not as the justification of the central tangent-cone or stationarity results. The CDCS Slater condition is a genuine assumption; whether it holds for all listed examples is a correctness/verification concern rather than a circularity. Overall, the derivation chain is not circular.

Assumptions & free parameters 3 free parameters · 7 assumptions · 0 invented entities

The central claim rests on the favorably structured CDCS class, a Slater-type constraint qualification, and standard variational-analysis machinery. Proposition 4 additionally requires smoothness and regularity conditions that are only partially stated. No new physical or mathematical entities are postulated; the cushioned linearization set is an algorithmic construction, not an invented entity.

free parameters (3)

μ (proximal coefficient) = μ > (L_f+L_g)β_θ (Theorem 2)
Chosen by the user or via backtracking; the descent analysis and all bounds depend on it.
δ (inner-loop tolerance)
Stopping threshold for the inner proximal-linear loop; appears in the local-regret and iteration-complexity bounds.
w (smoothing window)
Window length for the time-smoothed objective; controls the local-regret rate O(T/w^2) and the number of inner iterations.

assumptions (7)

domain assumption Favorable CDCS condition (Definition 1): f,g convex; f L_f-Lipschitz; g L_g-Lipschitz and differentiable; θ β_θ-smooth with bounded Jacobian; range bounded on the relevant set.
Defines the problem class and supplies all constants used in the bounds; if g is not differentiable or θ is not smooth, the proximal-linear construction fails.
domain assumption CDCS Slater CQ at active boundary points (Definition 2, eq. (28)): ∃d0 with φ'(ζ(¯x);Jζ(¯x)d0)<∇ψ(ζ(¯x))^T Jζ(¯x)d0.
Necessary for Theorem 1's tangent-cone equality and for Proposition 2's stationarity certificate; not verified for the examples when multiple constraints are active.
domain assumption ζ(¯x)∈int(dom φ) at active boundary points.
Used in Theorem 1 to ensure φ is locally Lipschitz and Hadamard directionally differentiable at ζ(¯x).
domain assumption Uniform parameters over all rounds (Assumption 1).
Allows the time-smoothed averages and regret/iteration bounds to use uniform constants L_f,L_g,β_θ,L_θ,M.
domain assumption For Proposition 4: existence of KKT multipliers, κ_f/κ_ϕ Lipschitz subgradient continuity (eqs. (78)-(80)), an η>0 with H_x(u_x)≤−η, and λ>0⇒x_+∈∂X.
These 'suitable conditions' control the Lagrange multiplier and the gap between linearized and true subgradients; they are substantial and not implied by Assumption 1.
ad hoc to paper Unstated β_g and β_ψ smoothness of ∇g_s and ∇ψ in Proposition 4's proof.
The proof bounds ||∇g_s(θ_s(x_+))−∇g_s(θ_s(x))||≤L_θ^2 β_g||d|| and ||∇ψ(ζ(x_+))−∇ψ(ζ(x))||≤L_ζ^2 β_ψ||d||, but β_g and β_ψ are not part of Definition 1, Assumption 1, or the proposition statement.
standard math Standard variational-analysis results: Rockafellar-Wets Thm 10.6, Mordukhovich Prop 1.107, Bonnans-Shapiro Prop 2.47, tangent/polar cone calculus.
Used throughout the proofs without reproving; these are standard background results in the field.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Online Optimization of Difference-of-Convex Compositions with Smooth Mappings." pith.science (2026). https://pith.science/paper/WT5J4URP

@misc{pith2026260719553,
  author       = {Pith},
  title        = {Pith review of: Online Optimization of Difference-of-Convex Compositions with Smooth Mappings},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WT5J4URP}},
  note         = {Machine review of arXiv:2607.19553}
}

read the original abstract

We study online optimization for a broad class of structured non-convex non-smooth problems where each loss is a composition of a difference-of-convex function with a smooth mapping, and the feasible region is defined by constraint functions of the same kind. We propose a time-smoothed proximal linear algorithm and a local-regret measure based on a proximal residual mapping. We show that this residual is a proper stationarity measure for the original problem: its fixed-point condition implies first-order stationarity. Our analysis relies on a tangent-cone characterization for a feasible region described by composite difference-of-convex constraints, which is of independent interest and allows each update to be computed via a convex optimization oracle, despite the non-convexity of the problem. We establish a local-regret bound and a bound on the total number of inner convex subproblems. We also derive an error bound connecting the proximal residual to the distance to stationarity, providing a quantitative certificate of approximate stationarity.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

106 extracted references · 4 linked inside Pith

[1]

Foundations and Trends

Introduction to online convex optimization , author=. Foundations and Trends. 2016 , publisher=

2016
[2]

, author=

X-Armed Bandits. , author=. Journal of Machine Learning Research , volume=
[3]

Advances in Neural Information Processing Systems , volume=

Nearly tight bounds for the continuum-armed bandit problem , author=. Advances in Neural Information Processing Systems , volume=
[4]

Proceedings of the fortieth annual ACM symposium on Theory of computing , pages=

Multi-armed bandits in metric spaces , author=. Proceedings of the fortieth annual ACM symposium on Theory of computing , pages=
[5]

Advances in Neural Information Processing Systems , volume=

Online convex optimization with stochastic constraints , author=. Advances in Neural Information Processing Systems , volume=
[6]

Advances in Neural Information Processing Systems , volume=

Online Learning under Adversarial Nonlinear Constraints , author=. Advances in Neural Information Processing Systems , volume=
[7]

International Conference on Machine Learning , pages=

Adaptive algorithms for online convex optimization with long-term constraints , author=. International Conference on Machine Learning , pages=. 2016 , organization=

2016
[8]

Advances in Neural Information Processing Systems , volume=

Online convex optimization for cumulative constraints , author=. Advances in Neural Information Processing Systems , volume=

Show all 106 references

[9]

IEEE Transactions on automatic control , volume=

Online convex optimization with time-varying constraints and bandit feedback , author=. IEEE Transactions on automatic control , volume=. 2018 , publisher=

2018
[10]

SIAM Journal on Optimization , volume=

Convergence analysis of a proximal-like minimization algorithm using Bregman functions , author=. SIAM Journal on Optimization , volume=. 1993 , publisher=

1993
[11]

International Conference on Machine Learning , pages=

The hedge algorithm on a continuum , author=. International Conference on Machine Learning , pages=. 2015 , organization=

2015
[12]

Algorithmic Learning Theory , pages=

Online non-convex learning: Following the perturbed leader is optimal , author=. Algorithmic Learning Theory , pages=. 2020 , organization=

2020
[13]

Joint european conference on machine learning and knowledge discovery in databases , pages=

Online learning in adversarial lipschitz environments , author=. Joint european conference on machine learning and knowledge discovery in databases , pages=. 2010 , organization=

2010
[14]

Advances in Neural Information Processing Systems , volume=

Online non-convex learning in dynamic environments , author=. Advances in Neural Information Processing Systems , volume=
[15]

Transactions on Machine Learning Research , year=

Dynamic Regret Analysis of Safe Distributed Online Optimization for Convex and Non-convex Problems , author=. Transactions on Machine Learning Research , year=
[16]

International Conference on Machine Learning , pages=

Efficient regret minimization in non-convex games , author=. International Conference on Machine Learning , pages=. 2017 , organization=

2017
[18]

International Conference on Machine Learning , pages=

Regret minimization in stochastic non-convex learning via a proximal-gradient approach , author=. International Conference on Machine Learning , pages=. 2021 , organization=

2021
[19]

Advances in neural information processing systems , volume=

Dynamic local regret for non-convex online forecasting , author=. Advances in neural information processing systems , volume=
[20]

2025 , month = feb, note =

Nadav Hallak , title =. 2025 , month = feb, note =

2025
[21]

The Thirty Sixth Annual Conference on Learning Theory , pages=

Online nonconvex optimization with limited instantaneous oracle feedback , author=. The Thirty Sixth Annual Conference on Learning Theory , pages=. 2023 , organization=

2023
[22]

, author=

Some NP-complete problems in quadratic and nonlinear programming. , author=. Mathematical programming , volume=
[23]

IEEE Transactions on signal processing , volume=

Sparse reconstruction by separable approximation , author=. IEEE Transactions on signal processing , volume=. 2009 , publisher=

2009
[24]

The Annals of Statistics , volume=

NEARLY UNBIASED VARIABLE SELECTION UNDER MINIMAX CONCAVE PENALTY , author=. The Annals of Statistics , volume=
[25]

Mechanical Systems and Signal Processing , volume=

Sparse reconstruction for blade tip timing signal using generalized minimax-concave penalty , author=. Mechanical Systems and Signal Processing , volume=. 2021 , publisher=

2021
[26]

Linear algebra and its applications , volume=

Computing the numerical radius , author=. Linear algebra and its applications , volume=. 1996 , publisher=

1996
[27]

Digital signal processing , volume=

Vector minimax concave penalty for sparse representation , author=. Digital signal processing , volume=. 2018 , publisher=

2018
[28]

IMA Journal of Numerical Analysis , volume=

An algorithm for computing the numerical radius , author=. IMA Journal of Numerical Analysis , volume=. 1997 , publisher=

1997
[29]

European Journal of Operational Research , volume=

DC approximation approaches for sparse optimization , author=. European Journal of Operational Research , volume=. 2015 , publisher=

2015
[30]

Journal of Convex Analysis , volume=

Filling the gap between lower-C1 and lower-C2 functions , author=. Journal of Convex Analysis , volume=
[31]

arXiv preprint arXiv:1605.00125 , year=

An accelerated algorithm for minimizing convex compositions , author=. arXiv preprint arXiv:1605.00125 , year=

arXiv
[32]

Mathematical Programming , volume=

Efficiency of minimizing compositions of convex functions and smooth maps , author=. Mathematical Programming , volume=. 2019 , publisher=

2019
[33]

2009 , publisher=

Variational analysis , author=. 2009 , publisher=

2009
[34]

Variational Analysis and Generalized Differentiation. I. Basic Theory, II. Applications. , author=. 2009 , publisher=

2009
[35]

Pacific Journal of mathematics , volume=

Minimization of functions having Lipschitz continuous first partial derivatives , author=. Pacific Journal of mathematics , volume=. 1966 , publisher=

1966
[36]

SIAM journal on imaging sciences , volume=

A fast iterative shrinkage-thresholding algorithm for linear inverse problems , author=. SIAM journal on imaging sciences , volume=. 2009 , publisher=

2009
[37]

Mathematical Programming , volume=

Descent methods for composite nondifferentiable optimization problems , author=. Mathematical Programming , volume=. 1985 , publisher=

1985
[38]

International Journal of Systems Science , volume=

A generalized proximal point algorithm for certain non-convex minimization problems , author=. International Journal of Systems Science , volume=. 1981 , publisher=

1981
[39]

Mathematical Programming , volume=

Composite proximal bundle method , author=. Mathematical Programming , volume=. 2013 , publisher=

2013
[40]

IEEE Transactions on Control of Network Systems , volume=

Dynamic regret bounds for constrained online nonconvex optimization based on polyak--lojasiewicz regions , author=. IEEE Transactions on Control of Network Systems , volume=. 2022 , publisher=

2022
[41]

Acta mathematica vietnamica , volume=

Convex analysis approach to DC programming: theory, algorithms and applications , author=. Acta mathematica vietnamica , volume=
[42]

Generalized differentiability/duality and optimization for problems dealing with differences of convex functions , author=. Convexity and Duality in Optimization: Proceedings of the Symposium on Convexity and Duality in Optimization Held at the University of Groningen, The Net...

1984
[43]

Applied mathematics and computation , volume=

On the difference of two maximal monotone operators: Regularization and algorithmic approaches , author=. Applied mathematics and computation , volume=. 2008 , publisher=

2008
[44]

SIAM Journal on Optimization , volume=

Convergence of new inertial proximal methods for DC programming , author=. SIAM Journal on Optimization , volume=. 2008 , publisher=

2008
[45]

Nonlinear Analysis and Optimization , pages=

Global minimization of a difference of two convex functions , author=. Nonlinear Analysis and Optimization , pages=. 2009 , publisher=

2009
[46]

2021 , publisher=

Modern nonconvex nondifferentiable optimization , author=. 2021 , publisher=

2021
[47]

Mathematics of Operations Research , volume=

Minimizing compositions of differences-of-convex functions with smooth mappings , author=. Mathematics of Operations Research , volume=. 2024 , publisher=

2024
[48]

Operations Research , volume=

Wasserstein Distributionally Robust Optimization and Variation Regularization , author=. Operations Research , volume=. 2024 , publisher=

2024
[49]

Computing

Tian, Lai and So, Anthony Man-Cho , booktitle=. Computing. 2022 , publisher=

2022
[51]

Mathematics of Operations Research , volume=

Computing B-stationary points of nonsmooth DC programs , author=. Mathematics of Operations Research , volume=. 2017 , publisher=

2017
[52]

Mathematics of Operations Research , volume=

Penalty and augmented Lagrangian methods for constrained DC programming , author=. Mathematics of Operations Research , volume=. 2022 , publisher=

2022
[53]

SIAM Journal on Optimization , volume=

Proximally guided stochastic subgradient method for nonsmooth, nonconvex problems , author=. SIAM Journal on Optimization , volume=. 2019 , publisher=

2019
[54]

SIAM Journal on optimization , volume=

Robust stochastic approximation approach to stochastic programming , author=. SIAM Journal on optimization , volume=. 2009 , publisher=

2009
[55]

Mathematical Programming Computation , volume=

A stochastic approximation method for approximating the efficient frontier of chance-constrained nonlinear programs , author=. Mathematical Programming Computation , volume=. 2021 , publisher=

2021
[56]

Computational Optimization and Applications , volume=

Algorithms for stochastic optimization with function or expectation constraints , author=. Computational Optimization and Applications , volume=. 2020 , publisher=

2020
[57]

Mathematics of Operations Research , volume=

Stochastic approximation proximal method of multipliers for convex stochastic programming , author=. Mathematics of Operations Research , volume=. 2023 , publisher=

2023
[58]

arXiv preprint arXiv:2002.05826 , year=

Statistical learning with conditional value at risk , author=. arXiv preprint arXiv:2002.05826 , year=

2002 arXiv
[59]

2013 , publisher=

Perturbation analysis of optimization problems , author=. 2013 , publisher=

2013
[60]

Set-Valued and Variational Analysis , volume=

Nonconvex and nonsmooth approaches for affine chance-constrained stochastic programs , author=. Set-Valued and Variational Analysis , volume=. 2022 , publisher=

2022
[61]

Statistica Sinica , pages=

Waveshrink with firm shrinkage , author=. Statistica Sinica , pages=. 1997 , publisher=

1997
[62]

Journal of computational and graphical statistics , volume=

Wavelet shrinkage denoising using the non-negative garrote , author=. Journal of computational and graphical statistics , volume=. 1998 , publisher=

1998
[63]

Minimization of functions having lipschitz continuous first partial derivatives

Larry Armijo. Minimization of functions having lipschitz continuous first partial derivatives. Pacific Journal of mathematics, 16 0 (1): 0 1--3, 1966

1966
[64]

A local regret in nonconvex online learning

Sergul Aydore, Lee Dicker, and Dean Foster. A local regret in nonconvex online learning. arXiv preprint arXiv:1811.05095, 2018

2018 arXiv
[65]

Dynamic local regret for non-convex online forecasting

Sergul Aydore, Tianhao Zhu, and Dean P Foster. Dynamic local regret for non-convex online forecasting. Advances in neural information processing systems, 32, 2019

2019
[66]

A fast iterative shrinkage-thresholding algorithm for linear inverse problems

Amir Beck and Marc Teboulle. A fast iterative shrinkage-thresholding algorithm for linear inverse problems. SIAM journal on imaging sciences, 2 0 (1): 0 183--202, 2009

2009
[67]

Perturbation analysis of optimization problems

J Fr \'e d \'e ric Bonnans and Alexander Shapiro. Perturbation analysis of optimization problems. Springer Science & Business Media, 2013

2013
[68]

X-armed bandits

S \'e bastien Bubeck, R \'e mi Munos, Gilles Stoltz, and Csaba Szepesv \'a ri. X-armed bandits. Journal of Machine Learning Research, 12 0 (5), 2011

2011
[69]

Descent methods for composite nondifferentiable optimization problems

James V Burke. Descent methods for composite nondifferentiable optimization problems. Mathematical Programming, 33: 0 260--279, 1985

1985
[70]

Dynamic regret analysis of safe distributed online optimization for convex and non-convex problems

Ting-Jui Chang, Sapana Chaudhary, Dileep Kalathil, and Shahin Shahrampour. Dynamic regret analysis of safe distributed online optimization for convex and non-convex problems. Transactions on Machine Learning Research, 2023

2023
[71]

Modern nonconvex nondifferentiable optimization

Ying Cui and Jong-Shi Pang. Modern nonconvex nondifferentiable optimization. SIAM, 2021

2021
[72]

A generalized proximal point algorithm for certain non-convex minimization problems

Masao Fukushima and Hisashi Mine. A generalized proximal point algorithm for certain non-convex minimization problems. International Journal of Systems Science, 12 0 (8): 0 989--1000, 1981

1981
[73]

Wavelet shrinkage denoising using the non-negative garrote

Hong-Ye Gao. Wavelet shrinkage denoising using the non-negative garrote. Journal of computational and graphical statistics, 7 0 (4): 0 469--488, 1998

1998
[74]

Waveshrink with firm shrinkage

Hong-Ye Gao and Andrew G Bruce. Waveshrink with firm shrinkage. Statistica Sinica, pages 855--874, 1997

1997
[75]

Kleywegt

Rui Gao, Xi Chen, and Anton J. Kleywegt. Wasserstein distributionally robust optimization and variation regularization. Operations Research, 72 0 (3): 0 1177--1191, 2024. doi:10.1287/opre.2022.2383

2024
[76]

Online nonconvex optimization with limited instantaneous oracle feedback

Ziwei Guan, Yi Zhou, and Yingbin Liang. Online nonconvex optimization with limited instantaneous oracle feedback. In The Thirty Sixth Annual Conference on Learning Theory, pages 3328--3355. PMLR, 2023

2023
[77]

A sound local regret methodology for online nonconvex composite optimization, February 2025

Nadav Hallak. A sound local regret methodology for online nonconvex composite optimization, February 2025. URL https://optimization-online.org/2025/02/a-sound-local-regret-methodology-for-online-nonconvex-composite-optimization/. Optimization Online, published February 17, 2025

2025
[78]

Regret minimization in stochastic non-convex learning via a proximal-gradient approach

Nadav Hallak, Panayotis Mertikopoulos, and Volkan Cevher. Regret minimization in stochastic non-convex learning via a proximal-gradient approach. In International Conference on Machine Learning, pages 4008--4017. PMLR, 2021

2021
[79]

Efficient regret minimization in non-convex games

Elad Hazan, Karan Singh, and Cyril Zhang. Efficient regret minimization in non-convex games. In International Conference on Machine Learning, pages 1433--1441. PMLR, 2017

2017
[80]

Introduction to online convex optimization

Elad Hazan et al. Introduction to online convex optimization. Foundations and Trends in Optimization , 2 0 (3-4): 0 157--325, 2016

2016
[81]

An algorithm for computing the numerical radius

Chunyang He and GA Watson. An algorithm for computing the numerical radius. IMA Journal of Numerical Analysis, 17 0 (3): 0 329--342, 1997

1997
[82]

Generalized differentiability/duality and optimization for problems dealing with differences of convex functions

J-B Hiriart-Urruty. Generalized differentiability/duality and optimization for problems dealing with differences of convex functions. In Convexity and Duality in Optimization: Proceedings of the Symposium on Convexity and Duality in Optimization Held at the University of Groni...

1984
[83]

Nearly tight bounds for the continuum-armed bandit problem

Robert Kleinberg. Nearly tight bounds for the continuum-armed bandit problem. Advances in Neural Information Processing Systems, 17, 2004

2004
[84]

Multi-armed bandits in metric spaces

Robert Kleinberg, Aleksandrs Slivkins, and Eli Upfal. Multi-armed bandits in metric spaces. In Proceedings of the fortieth annual ACM symposium on Theory of computing, pages 681--690, 2008

2008
[85]

The hedge algorithm on a continuum

Walid Krichene, Maximilian Balandat, Claire Tomlin, and Alexandre Bayen. The hedge algorithm on a continuum. In International Conference on Machine Learning, pages 824--832. PMLR, 2015

2015
[86]

Dc approximation approaches for sparse optimization

Hoai An Le Thi, T Pham Dinh, Hoai Minh Le, and Xuan Thanh Vo. Dc approximation approaches for sparse optimization. European Journal of Operational Research, 244 0 (1): 0 26--46, 2015

2015
[87]

Minimizing compositions of differences-of-convex functions with smooth mappings

Hoai An Le Thi, Van Ngai Huynh, and Tao Pham Dinh. Minimizing compositions of differences-of-convex functions with smooth mappings. Mathematics of Operations Research, 49 0 (2): 0 1140--1168, 2024

2024
[88]

Penalty and augmented lagrangian methods for constrained dc programming

Zhaosong Lu, Zhe Sun, and Zirui Zhou. Penalty and augmented lagrangian methods for constrained dc programming. Mathematics of Operations Research, 47 0 (3): 0 2260--2285, 2022

2022
[89]

Online learning in adversarial lipschitz environments

Odalric-Ambrym Maillard and R \'e mi Munos. Online learning in adversarial lipschitz environments. In Joint european conference on machine learning and knowledge discovery in databases, pages 305--320. Springer, 2010

2010
[90]

Convergence of new inertial proximal methods for dc programming

Paul-Emile Maing \'e and Abdellatif Moudafi. Convergence of new inertial proximal methods for dc programming. SIAM Journal on Optimization, 19 0 (1): 0 397--413, 2008

2008
[91]

Variational Analysis and Generalized Differentiation

Boris Mordukhovich. Variational Analysis and Generalized Differentiation. I. Basic Theory, II. Applications. Springer, 2009

2009
[92]

On the difference of two maximal monotone operators: Regularization and algorithmic approaches

Abdellatif Moudafi. On the difference of two maximal monotone operators: Regularization and algorithmic approaches. Applied mathematics and computation, 202 0 (2): 0 446--452, 2008

2008
[93]

Dynamic regret bounds for constrained online nonconvex optimization based on polyak--lojasiewicz regions

Julie Mulvaney-Kemp, SangWoo Park, Ming Jin, and Javad Lavaei. Dynamic regret bounds for constrained online nonconvex optimization based on polyak--lojasiewicz regions. IEEE Transactions on Control of Network Systems, 10 0 (2): 0 599--611, 2022

2022
[94]

Some np-complete problems in quadratic and nonlinear programming

Katta G Murty, Santosh N Kabadi, et al. Some np-complete problems in quadratic and nonlinear programming. Mathematical programming, 39 0 (2): 0 117--129, 1987

1987
[95]

Computing b-stationary points of nonsmooth dc programs

Jong-Shi Pang, Meisam Razaviyayn, and Alberth Alvarado. Computing b-stationary points of nonsmooth dc programs. Mathematics of Operations Research, 42 0 (1): 0 95--118, 2017

2017
[96]

Variational analysis, volume 317

R Tyrrell Rockafellar and Roger J-B Wets. Variational analysis, volume 317. Springer Science & Business Media, 2009

2009
[97]

Composite proximal bundle method

Claudia Sagastiz \'a bal. Composite proximal bundle method. Mathematical Programming, 140 0 (1): 0 189--233, 2013

2013
[98]

Online non-convex learning: Following the perturbed leader is optimal

Arun Sai Suggala and Praneeth Netrapalli. Online non-convex learning: Following the perturbed leader is optimal. In Algorithmic Learning Theory, pages 845--861. PMLR, 2020

2020
[99]

Convex analysis approach to dc programming: theory, algorithms and applications

Pham Dinh Tao and LT Hoai An. Convex analysis approach to dc programming: theory, algorithms and applications. Acta mathematica vietnamica, 22 0 (1): 0 289--355, 1997

1997
[100]

Computing D -stationary points of -margin loss SVM

Lai Tian and Anthony Man-Cho So. Computing D -stationary points of -margin loss SVM . In Proceedings of the 25th International Conference on Artificial Intelligence and Statistics, volume 151 of Proceedings of Machine Learning Research, pages 3772--3793. PMLR, 2022. URL https:...

2022
[101]

Global minimization of a difference of two convex functions

Hoang Tuy. Global minimization of a difference of two convex functions. In Nonlinear Analysis and Optimization, pages 150--182. Springer, 2009

2009
[102]

Vector minimax concave penalty for sparse representation

Shibin Wang, Xuefeng Chen, Weiwei Dai, Ivan W Selesnick, Gaigai Cai, and Benjamin Cowen. Vector minimax concave penalty for sparse representation. Digital signal processing, 83: 0 165--179, 2018

2018
[103]

Computing the numerical radius

GA Watson. Computing the numerical radius. Linear algebra and its applications, 234: 0 163--172, 1996

1996
[104]

Sparse reconstruction by separable approximation

Stephen J Wright, Robert D Nowak, and M \'a rio AT Figueiredo. Sparse reconstruction by separable approximation. IEEE Transactions on signal processing, 57 0 (7): 0 2479--2493, 2009

2009
[105]

Sparse reconstruction for blade tip timing signal using generalized minimax-concave penalty

Jinghui Xu, Baijie Qiao, Junjiang Liu, Chunyan Ao, Guangrong Teng, and Xuefeng Chen. Sparse reconstruction for blade tip timing signal using generalized minimax-concave penalty. Mechanical Systems and Signal Processing, 161: 0 107961, 2021

2021
[106]

Online non-convex learning in dynamic environments

Zhipan Xu and Lijun Zhang. Online non-convex learning in dynamic environments. Advances in Neural Information Processing Systems, 37: 0 51930--51962, 2024

2024
[107]

Nearly unbiased variable selection under minimax concave penalty

Cun-Hui Zhang. Nearly unbiased variable selection under minimax concave penalty. The Annals of Statistics, 38 0 (2): 0 894--942, 2010

2010
[108]

Nonconvex composite functional constraints via first-order augmented lagrangian methods under local regularity, July 2026

Linglingzhi Zhu and Jiajin Li. Nonconvex composite functional constraints via first-order augmented lagrangian methods under local regularity, July 2026. URL https://arxiv.org/abs/2607.08954. arXiv preprint arXiv:2607.08954

2026 arXiv

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.

[1] [1]

Foundations and Trends

Introduction to online convex optimization , author=. Foundations and Trends. 2016 , publisher=

2016

[2] [2]

, author=

X-Armed Bandits. , author=. Journal of Machine Learning Research , volume=

[3] [3]

Advances in Neural Information Processing Systems , volume=

Nearly tight bounds for the continuum-armed bandit problem , author=. Advances in Neural Information Processing Systems , volume=

[4] [4]

Proceedings of the fortieth annual ACM symposium on Theory of computing , pages=

Multi-armed bandits in metric spaces , author=. Proceedings of the fortieth annual ACM symposium on Theory of computing , pages=

[5] [5]

Advances in Neural Information Processing Systems , volume=

Online convex optimization with stochastic constraints , author=. Advances in Neural Information Processing Systems , volume=

[6] [6]

Advances in Neural Information Processing Systems , volume=

Online Learning under Adversarial Nonlinear Constraints , author=. Advances in Neural Information Processing Systems , volume=

[7] [7]

International Conference on Machine Learning , pages=

Adaptive algorithms for online convex optimization with long-term constraints , author=. International Conference on Machine Learning , pages=. 2016 , organization=

2016

[8] [8]

Advances in Neural Information Processing Systems , volume=

Online convex optimization for cumulative constraints , author=. Advances in Neural Information Processing Systems , volume=